The present invention is concerned with computer systems, and is more particularly concerned with error detection by bus bridge circuits in computer systems.
FIG. 1 is a block diagram of a typical computer system which employs bus bridge circuits (xe2x80x9cbus bridgesxe2x80x9d or xe2x80x9cbridgesxe2x80x9d) to provide interconnections between buses. Reference numeral 10 generally indicates the computer system. The computer system 10 includes a processor 12 connected via a system bus 14 to a system memory 16. A host bridge 18 connects the system bus 14 to a first PCI (Peripheral Component Interconnect) bus 20. A first PCI/PCI bridge 22 connects the first PCI bus 20 to a second PCI bus 24. A second PCI/PCI bridge 26 connects the second PCI bus 24 to a third PCI bus 28. I/O (input/output) adapters 30 are connected to the third PCI bus 28.
A third PCI/PCI bridge 32 connects the second PCI bus 24 to a fourth PCI bus 34. I/O adapters 36 are connected to the fourth PCI bus 34. A fourth PCI/PCI bridge 38 connects the first PCI bus 20 to a fifth PCI bus 40. I/O adapters 42 are connected to the fifth PCI bus 40.
FIG. 2 shows additional details of the portion of FIG. 1 that is concerned with the fourth PCI/PCI bridge 38 and the fifth PCI bus 40. In particular, FIG. 2 shows that the fourth PCI/PCI bridge 38 stores base address data 44, limit address data 46, secondary bus number data 48, and subordinate bus number data 50. As will be appreciated by those who are skilled in the art, corresponding data is also stored in the first PCI/PCI bridge 22, the second PCI/PCI bridge 26 and the third PCI/PCI bridge 32. In accordance with conventional practice in connection with the well-known PCI protocol, the base address data 44 and the limit address data 46 define a range of addresses to which request commands may be directed. Also in accordance with conventional practice, the secondary bus number data 48 and the subordinate bus number data 50 together define a range of bus numbers to which completions may be directed.
The well-known PCI bus protocol is an industry standard administered by the PCI Special Interest Group. Also available from PCI Special Interest Group for a fee is information regarding a recent extension of the PCI protocol, referred to as xe2x80x9cPCI-Xxe2x80x9d and described in xe2x80x9cPCI-X Addendum PCI Local Bus Specification Revision 1.0a Jul. 24, 2000xe2x80x9d.
One new feature provided in PCI-X is known as the xe2x80x9csplit operationxe2x80x9d. To briefly illustrate a split operation in the context of a read command, a requesting device (xe2x80x9crequesterxe2x80x9d) issues a read command, and the target device (xe2x80x9ccompleterxe2x80x9d) responds with a split response. The completer later issues a split completion to provide the data originally requested by the requester. The split operation feature was added to save bandwidth on the PCI bus by avoiding retries.
FIG. 3 is a flow chart that illustrates performance in accordance with conventional practices of a split read operation in a case where a bridge is positioned between the requesting device and the ultimate target.
The process of FIG. 3 starts at 52, and proceeds to block 54. At block 54 the bridge receives a read command from the requester. The received command includes a PCI address that identifies the target, and an xe2x80x9cattributexe2x80x9d which indicates the return address (bus number) for the requester. Following block 54 is block 56. At block 56 the bridge tests the PCI address received at block 54 against the address range defined by the base address data and the limit address data stored in the bridge to determine whether the read command is properly routable by the bridge. Assuming that the received PCI address is proper, block 58 then follows. At block 58 the bridge accepts the read command. Then, at block 60, the bridge issues a split response to the requester.
Following block 60 is block 62. At block 62 the bridge issues a read command on its other bus, passing along the PCI address and the attribute which were received at block 54. The bridge then receives a split response from the completer (block 54). After a certain period of time the bridge then receives a split completion from the completer. The split completion includes an address, i.e. a bus number, used to route the completion back to the requester. Following block 66 is decision block 68. At decision block 68 the bridge determines whether the bus number included in the split completion matches the range of bus numbers supported by the bridge (i.e., the range of bus numbers defined by the secondary bus number data and the subordinate bus number data). If the bus number does match the supported range, then the bridge completes the read operation by transmitting the split completion to the requester (block 70) and the process ends (72).
If a negative determination is made at decision block 68, i.e. if the bus number received in the split completion does not match the range of bus numbers supported by the bridge, then block 74 follows decision block 68. At block 74 the bridge ignores (master aborts) the split completion and the process ends (72).
The path through block 74, leading to aborting of the split completion, may occur if the attribute originally provided by the requester was incorrect. This, in turn, may result from corruption of data in the requester, or reassignment of bus numbers by the host without the requester having been properly updated.
The present inventors have recognized certain problems that arise in connection with the conventional split-operation handling process of FIG. 3, and particularly with the aborting of the split completion at block 74. Specifically, the aborting of the split completion leads the completer to issue a general error message which does not (cannot) identify the requester, which is likely the source of the bad bus number. As a result, the system may be unable to identify the source of the malfunction, and may be prevented from making a recovery from the error condition. Furthermore, upon the issuance of the split response at block 60, the requester may lock resources, such as a buffer in which the desired data is to be written. Also, the bridge may lock a resource, such as a register for storing state information, upon receipt of the split response at block 64. Both the locked resources at the requester and the locked resources at the bridge may remain locked indefinitely upon aborting of the split completion.
According to a first aspect of the invention, a method of handling a split operation at a bus bridge is provided. The inventive method includes receiving at the bus bridge a read or write command from a requesting device, where the command includes a bus number for routing a completion of the command. The inventive method further includes comparing the bus number received from the requesting device with the return route bus number range of the bus bridge, issuing a split response to the requesting device if the bus number matches the return route bus number range of the bus bridge, and aborting the command if the bus number does not match the return route bus number range of the bus bridge.
In at least one embodiment, the read or write command may be received at the bus bridge via a PCI-X bus. The requesting device may be an I/O adapter.
According to a second aspect of the invention, a computer system is provided. The inventive computer system includes a requesting device connected to a first bus, a completing device connected to a second bus, and a bus bridge interconnecting the first bus and the second bus. The bus bridge is operative to receive a read or write command from the requesting device, where the command includes a bus number for routing a completion of the command. The bus bridge is further operative to (a) compare the bus number with the return route bus number range of the bus bridge, (b) issue a split response to the requesting device if the bus number matches the return route bus number range of the bus bridge, and (c) abort the command if the bus number does not match the return route bus number range. The completing device may be a host bridge.
Numerous other aspects are provided as are computer program products. Each inventive computer program product may be carried by a medium readable by a computer (e.g., a carrier wave signal, a floppy disk, a hard drive, a random access memory, etc.).
With the present invention, an error in an attribute (return address bus number) included by a requesting device in a read or write command can be identified early in the split operation process, so that a requesting device which causes a malfunction can be identified, thereby enhancing the likelihood that the system can recover from the malfunction. Furthermore, the error is detected early enough to prevent open-ended locking of resources of the requesting device and the bridge.
Other objects, features and advantages of the present invention will become more fully apparent from the following detailed description of exemplary embodiments, the appended claims and the accompanying drawings.